A survey on learning-based dense 3D reconstruction

PhD Qualifying Examination


Title: "A survey on learning-based dense 3D reconstruction"

by

Mr. Jingyang ZHANG


Abstract:

Dense reconstruction takes overlapping images with known cameras and 
produces complete surface, which is the last stage of typical 3D 
reconstruction pipeline. One of the most popular dense reconstruction 
methods is multi-view stereo (MVS) which estimates depth maps for every 
image and fuses them into final surface. In recent years, deep learning 
techniques have brought significant improvement to MVS. And we expect the 
great potential to transfer this success to other dense methods with 
alternative workflow.

In this survey, we first review the development of learning-based stereo 
methods. Then we identify the visibility issue and memory issue of these 
methods and introduce possible solutions. Finally, we analyze some related 
tasks and discuss other possible workflow of learning-based dense 
reconstruction.


Date:			Monday, 28 December 2020

Time:                  	2:00pm - 4:00pm

Zoom meeting: 
https://hkust.zoom.us/j/7272663214?pwd=VzNmY2xIL2w5LzdObTNsd29LRDg2UT09

Committee Members:	Prof. Long Quan (Supervisor)
 			Dr. Qifeng Chen (Chairperson)
 			Prof. Pedro Sander
 			Dr. Dan Xu


**** ALL are Welcome ****